Geosynchronous Satellite Maneuver Classification via Supervised Machine Learning

Thomas G. Roberts, Massachusetts Institute of Technology; Richard Linares, Massachusetts Institute of Technology

Keywords: Detection Algorithms, Satellite Maneuver Detection, Geosynchronous Orbit, Geostationary Orbit, Satellite Patterns of Life, Machine Learning, Convolutional Neural Networks, Two-Line Element Sets

Abstract:

This work describes an approach for detecting the components of longitudinal shift maneuvers in the geosynchronous (GEO) orbital regime using convolutional neural networks trained on publicly available two-line element (TLE) data from the U.S. Space Command’s space object catalog. A method for converting TLE data to geographic position histories—longitude, latitude, and altitude positions over time in the Earth-centered, Earth-fixed geographic reference frame—and labeling longitudinal shift maneuvers by inspection is described. A preliminary maneuver detection algorithm is designed, trained, and tested on all GEO satellites in orbit from January 1 to December 31, 2020. Performance metrics are presented for a suite of algorithms trained on data sets corresponding to ten years’ worth of geographic position time-histories labeled with longitudinal shift maneuvers. When detected, longitudinal shift maneuvers can be used to identify anomalous behavior in GEO. In this work, a satellite’s behavior is considered nominal if it adheres to the satellite’s pattern of life (PoL)—its previous on-orbit behavior made up of sequences of both natural and non-natural behavioral modes, including routine station-keeping, other on-orbit maneuvers, and uncontrolled motion—and anomalous if it deviates from the satellite’s PoL. Identifying anomalous satellite behavior is of critical interest to space situational awareness system operators, who may choose to task their sensors to obtain more observations of anomalous behavior, and satellite operators themselves, who may wish to diagnose its root cause. Applications of this work for international space policymaking is also discussed.

Date of Conference: September 14-17, 2021

Track: Machine Learning for SSA Applications

View Paper